Flickr Search Hack Powered by Mouse-Made Doodles

Carl Bialik from WSJ writes "Retrievr gives budding artists an impractical but addictive way to find photographs on Flickr: a search engine powered exclusively by mouse-made doodles. From the article: 'Retrievr, Mr. Langreiter says, "doesn't look at specific forms." Art history buffs might like to think of it as photo-search by way of Impressionism. The Retrievr engine dissects a photo like a gallery connoisseur who lost his bifocals: It focuses on regions of colors rather than specific shapes and lines. "It is, actually, a simple scheme," says Mr. Langreiter. Retrievr creates and stores a compact representation of each photo in its database. The system pulls only the most important features — broad shapes, blocks of color and spatial relationships between different colored areas — out of detailed images to create shorthand approximations of every photo. (The storage mechanism extracts the 120 "strongest" features from an image to create something called a "wavelet transform," which contains much less data than the photo itself and facilitates lightning-fast searches.)'"

Flickr Retrievr

Direct Link [systemone.at]

Requires Flash.

Re:Flickr Retrievr

It's powered by "mouse-made doodles" and apparently you're not doodling enough. :)

Re:Flickr Retrievr

FTA

In its current incarnation, Retrievr runs on a single computer.

Ow. The Slashdotting. It hurts.

That was quick

I think we just made the world record for the most number of boobies sketched out on the internet simultaneously.

FFS!

20 minutes of expert artistry
are there no nipples on flickr?

Combo of Retrievr & Online Dating

Hmmm... Think about the combination of this and online dating sites! Especially if I could upload a target photo instead of sketching! ... I think I have some old Cindy Crawford JPGs laying around here somewhere (*dream on*).

That was quick

Already partly slashdotted. Very slow and sometimes you don't get in.
But this is an interesting idea, fun if nothing else.
I drew a tree and I got a pineapple with a guy's face in it, a chinese guy standing in front of a gate, and a dragonfly. Maybe I need to brush up on my drawing skills.

*groan*

It's been done before

I read about this [imgseek.net] a little while ago. Same principle. It uses a Haar transform (for those unfamiliar with multimedia signal processing and wavelets, specifically, the Haar transform is a specific wavelet transform based on the Haar wavelet and the associated orthogonal basis). The idea is that you compare the low frequency component of an image to the low frequency component of a rough drawing (which is pretty low frequency to begin with) and they should be pretty close of the images have anything in common.

Re:It's been done before

Haar wavelet though? While it's easier computationally (since the mother/father wavelets are peicewise linear in the 1D case) I always saw it as being a "lesser" wavelet in the sense of compression/reconstruction quality and ability to discern edges/other dramatic changes in data

I'm just saying what imgSeek uses. It's certainly a very easy wavelet to implement via lifting. I think it's probably used because more complex wavelets wouldn't be of any help since the rough drawing is so rough to begin with. In the end you could probably do the same thing with a DCT. Wish I had time to experiment.

powered by mouse-made doodles

Sounds like a pretty high risk factor for Hanta virus infection :P

Applied to museums?

To find Van Goghs, draw a whirlpool.
To find Pollocks, draw a can of paint.
To find Warhols, draw four cans of paint.
To find modern art sculptures, throw the tablet against a wall.

Tits

Oops I made the search xplode

Other flickr Mashups

This is very old .. I read about this first on webmonkey in Feb.

Ten Best Flickr Mashups
by Michael Calore 24 Feb 2006

Here's the link: Ten Best Flickr Mashups [webmonkey.com]

Rating the doodles

Keep in mind that there's a rating system for the doodles also.. there's some pretty cool artwork in there, as well as 50% boobies, dicks and strange V shapes (everyone draws them a little different). Pretty fun, it's under the Art of Retrivr

But the real question is...

...how does this help me find pr0n?

Mouse-made what??

For a second there I read "Mouse-Made Doodies"...

Finally, a good use for surplus mouse poop!

Feature Vector

120 features get mapped into a feature vector, effectively pinpointing a position in 120 dimensional space. All of the other images are indexed in the space, and it's a simple nearest-neighbor search to find the best matches. The interesting thing here is that funky things happen to space when you are in very high dimensions, and without creative indexing, it may be just as quick to do a scan and compare against the whole database. Obviously, not optimal. That's what they mean by "simple", since some multimedia search systems deal with indexes of thousands of features - thousands of dimensions.

The MPAA can use this

Some kind of first-pass search system to find rough matches between Flickr/Youtube/etc posts and copyrighted material will be a big win for the MPAA. There are ways to align and compare pictures, but they're computationally expensive and compare two images, they don't do a general search. This thing might be usable as a first search used to find possible matches, which then get a more detailed examination by the expensive algorithm.

I failed!

I tried to draw boobs and got this instead http://www.flickr.com/photos/19406332@N00/41144732 [flickr.com]

I think it needs some bugs worked out. It searches as well as a search engine.

Wavelet Transforms

Interesting technique. I used things called Active Shape Models [man.ac.uk] in my MSc thesis which do pretty much the same kind of thing if I understand the article correctly. I've since lost touch with the academic research. Can anyone who still works in the area give me a quick run down of the differences and pros/cons?

Bob

arabs first

whatever I draw, this picture at http://www.flickr.com/photos/96302514@N00/58916441 / [flickr.com] is always in the list of results!

wavelet first

just for the record: the post claims that ``the storage mechanism extracts the 120 "strongest" features from an image to create something called a "wavelet transform"'' but this is quite misworded. Indeed if you look into the original research project, [washington.edu] you see that ``the algorithm performs a wavelet transform [wikipedia.org] on every image, and then collects just the few largest coefficients from this transform''.

Does it work? Yes!

But don't take my word for it; see for yourself! http://i11.tinypic.com/48bt477.png [tinypic.com]

Urg

(The storage mechanism extracts the 120 "strongest" features from an image to create something called a "wavelet transform," which contains much less data than the photo itself and facilitates lightning-fast searches.)'"

If you're going to simplify, atleast get it somewhere near correct. A wavelet transform doesn't extract features. Features is a human-made concept. A wavelet transform is simply a transform (or for this purpose, a very lossy compression algorithm), very similar to the Fourier Transform, except that it has locality which is why it performs soo much better on non-uniform data.

I mean, 'something called a wavelet transform'. A short explanation linking it Fourier might have been apt, but wavelets are hardly voodoo.

'facilitates lightning-fast searches'.. oohh, thanks for telling us. I would never have guessed that after transforming the data down to 12 vectors, searching would be a lot faster. I mean, if they actually had indexed the data in a clever way or something specifically to speed up searches, this sentence would have made sense.. but they just transformed it. It's not voodoo and market-speech is bad!

Google?

I wonder, is Google looking at these alternative search methods? He mentions in the article that to do this to the entire Flicker would take 'some sort of Google-like infrastructure'...
I used to assume that, being the current king of text search, Google would be expanding into more intelligent media searches, like this kind of 'similar image'-search, automatic tagging (like automatically indexing a picture 'nature' and 'winter'), and searching for songs by entering a couple of notes.
They never did though, and instead began diversifying with maps, mail and office packs. They've been talking about using fingerprint technology in Youtube, though, so I guess we could get some fallout from that.

Failure

Is it just me, or does this thing always find nothing that actually looks like the picture? I drew a blue sky, green grass background, and then a brown tree trunk with dark green leaves... Just all blobs, basically. Not a single one of the pictures had a tree in it. They didn't look ANYTHING like the 'picture' I drew.

Im Suprised

As usefull as a broken toaster. Only good porn hits will save thee

Oblig.

"What are we going to do tonight, Brain?"

"Same thing we do every night Pinky, try and take over the world! We shall create millions of doodles which will keep the world's intelligence services occupied long enough for us to take over without opposition."

Seems Like...

It seems like if I do a search by an uploaded file, and I upload (or point to the URL of) an image that I have posted on Flickr, that image should come up first in the list of results, but it doesn't. How are these other pictures better matches to my picture than that picture's duplicate?

Same Type of idea for 3D Models

For a 3D model search engine, princeton has provided a 3d model search engine based on sketches of the three axis aligned planes.

http://shape.cs.princeton.edu/search.html [princeton.edu]

an interesting note about color

(My group did a content based image search engine at OSU: http://www.cse.ohio-state.edu/~hakan/CIS772/index. html [ohio-state.edu])

The interesting thing about color is that it is good enough ("Voxel 4" was our best if you find our website) to find images that have been resized, run through *some* effects filters, or rotated.

So if you use the same usericon etc, resized, on unconnected websites, that could be used to correlate your identities once this feature is in a global search engine. Also, the clustering of related images, for efficiency, restricts your results to the same cluster (for our implementation), which may not be apparent at first.

p.s. Thanks #Wikipedia for the gigabytes of images URL.

Please tell me

Please tell me I'm not the only one who thought Mouse Doodles referred to excrement at first?

Wavelets and doodling searches

I worked for Pacific Press Service in Tokyo developing photo copyright and library tech until 94. I first saw a photograph search engine developed by Fujitsu around 92-93 I believe. It required the user to draw the type of image composition very roughly with a mouse and paintbox. So you would draw a horizon line, fill the bottom with blue and draw a yellow circle above if you wanted photos of the sea and sun. No wavelets at that time.

I then corresponded briefly with Ingrid Daubechies of AT&T who brought wavelets [beyonddiscovery.org] to the U.S., and was kind enough to send some of her papers. Wavelets are neat because it is like getting a paintbox full of different waveforms, localized as another poster mentions not just a fourier of the entire image. Anyway they are much better known now, so you can find it on the net.

This is not really the same as Barnsley's fractal compression [wikipedia.org] one startup worked on around that time IIRC. They basically had a library of fractals which would be matched to image features, and once you had covered the entire image with them you would be able to zoom into it infinitely, since fractals are self-similar. You wouldn't necessarily get new detail but it would fool you into thinking you were. (I wonder if they liscensed it to anyone). They claimed 400:1 compression, etc. I don't know if they were the basis of LivePicture or if that was wavelet based.

These technologies all have two things in common, which is selecting an algorithmic strategy for talking about images, and storing it so efficiently that the data can be found quickly. The old Fujitsu system ran on a NEWS workstation IIRC, and it was blisteringly fast compared to any system I have ever seen. Only problem is doodles all look pretty much the same unless you are talented and patient.

It seems PNI (Picture Network Interactive)'s natural language recognition text searching for photos was the best, it was just text but used software supposedly developed for the White House. Only thing was they wanted to take over the entire industry with online contracts (this was around 1993) so everyone hated them. Nice tech though.

Anyway, wavelets may not be the entire solution but certainly they are a very useful way to describe data (not just a photo) and undoubtedly have lots of potential applications that just haven't materialized yet. Here's some tidbits Lancaster's links [tinaja.com] ImgSeek [imgseek.net]
  Perl Haar decomposition and seeking [simon-cozens.org]
  Blitzwave lib [sourceforge.net]
  wvlt [cs.ubc.ca]
  wvlt #2 [wavelet.org]
  Wavelet.org [wavelet.org]
  WSQ used for FBI fingerprinting [cognaxon.com]

Energy source?

I don't see where the article mentions exactly how the "mouse-made doodies" are mined for the energy to power the hack. Possibly by way of methane extraction? Oh, "doodles", not "doodies". Never mind.

Re:He gone

Maximum Thread count reached.

Ah, the web site doesn't work, but at least we get really nice feeling sheets!